Czech Technical University in Prague, Klokner Institute

Competitive comparison of load

combination models
Milan Holicky and Miroslav Sykora
Czech Technical University in Prague,
Klokner Institute

Introduction
Comparison based on previous experience
Numerical example
Concluding remarks

1st International Symposium on Uncertainty Modelling in

Engineering, 2 - 3 May 2011, Prague, Czech Republic 1
 Introduction
• Civil engineering structures often exposed to combinations of
time-variant loads (climatic actions, imposed loads)

• Several load combination models applied in reliability studies

• The present study aimed at comparison of three selected

approaches:
- Rule proposed by Turkstra (1970)
- Rectangular wave renewal processes with fixed durations of
pulses, Ferry Borges & Castanheta (1971) – FBC models
- Rectangular wave renewal processes with random durations
between renewals and random durations of load pulses, Rackwitz
(1998) and Sykora (2005)

• Comparison based on previous experience, numerical study

2
 Basic assumptions
• Resistance, geometry variables, permanent actions and model
uncertainties - time-invariant
• Time-variant actions described by stationary, ergodic and regular
processes

durations of „on“-states Ton

Q(t)

q
“off”-state “on”-state

t
durations between renewals Tren
reference period tref

3
 Turkstra’s rule

t*
Q1 Q1,tref q1
Q1,tref

Q2 q2
Q2,t* Q2,t*

t* t* tref t fQ(q)

4
 FBC models

Q1 q1 Q1,tren1
Q1,tren1

Q2 tren1 q2
Q2,tren2
Q2,tren1
Q2,tren1
Q2,tren2
tren2 tref t ftren(q)

5
 Renewal processes
• Upper bound on the failure probability in most applications
(initial failure probability + outcrossing rate)
Ton1 Ton1

Q1
Q1

Ton2 Ton2
Tren1 Tren1 Tren1
Q2 Q2

tref Tren2 Tren2 t

6
 Comparison based on previous experience
• Applicability of reliability methods
(+) Turkstra - any of well-established methods for the time-
invariant analysis
(-) FBC models – Rackwitz-Fiessler algorithm available in few
software products
(-) Renewal processes – upper bound unavailable in software
products

• Accuracy
(0) Turkstra – sufficiently accurate in most cases (given the
leading action is identified)
(0) FBC models – exact solution (applicability to short-term
actions like storms and earthquakes disputable)
(0) Renewal processes – applicable for many types of actions,
crude approximation when time-invariant variables dominant
7
 Comparison based on previous experience
• Estimation of partial factors (calibration studies)
(+) Turkstra - straightforward
(-) FBC models – easy for time-invariant variables, difficulties for
time-variant loads
(0) Renewal processes – straightforward when a dominant load
case can be identified

• Non-stationary cases (out of the scope of the contribution)

(-) Turkstra and FBC models – upper bound (maximum load
effect and minimum resistance) may be overly conservative
(+) Renewal processes – efficient analysis using the Laplace
transform

8
 Numerical example
• Reliability analysis of low-rise frames exposed to snow and wind,
Schleich et al. (2002) and Sadovsky & Pales (2008)
• Design according to Eurocodes
• Models for the monthly maxima of the climatic loads -
meteorological data for six locations in Germany
• Snow present with the probability pon; wind always present

20 m
20 m
10 m

25 m 22 m 15 m
9
 Basis of analysis
• Limit state function: g[X(t)] = KRR - KE[G + S(t) + W(t)]
• Reference period – 50 years

Variable Dist. μX/xk VX pon,X

Resistance R LN 1.18 0.08 -
Permanent load G N 1 0.10 -
Snow on roof S (Münster) GU 0.26 1.17 0.23
Wind action W (Münster) GU 0.17 0.67 1
Resistance uncertainty KR LN 1.15 0.05 -
Load effect uncertainty KE LN 1.0 0.10 -

• Parameter - load ratio χ = (sk + wk) / (gk + sk + wk)

10
 Reliability index – frame A (χ = 0.8)
One dominant action (frame A – snow, frame C - wind)
6
β
5

Turkstra - wind
4
3.8

3 renewal processes
FBC Turkstra - snow

Braunlage
Münster

Bremen

Stuttgart
Aachen

Berlin
11
 Reliability index – frame B (χ = 0.8)
Comparable effects of snow and wind (frame B)
6
β
FBC Turkstra - wind
5
Turkstra - snow

4
3.8
renewal processes
3

Braunlage
Münster

Bremen

Stuttgart
Aachen

Berlin
12
Reliability index vs. χ – frame B, Berlin
6
β
Turkstra - wind
5 Turkstra - snow

FBC
3
renewal processes

2
0 0.2 0.4 0.6 0.8 χ 1

13
Partial factors γM0 and γG vs. χ – frame B, Berlin
(βt = 3.8)
3.5
γM0, γG
3

2.5

2 renewal processes
1.5 γG
1
γM0
0.5 Turkstra – snow ≈ FBC

0
0 0.2 0.4 0.6 0.8 χ 1

14
Partial factors γS and γW × ψW vs. χ – frame B, Berlin
(βt = 3.8)
3.5
γS, γW × ψW
3 γS
2.5

2
renewal processes
1.5 FBC
Turkstra – snow
1
γW × ψW
0.5

0
0 0.2 0.4 0.6 0.8 χ 1

15
 Conclusions
• Selection of a model for the load combination may be a key
issue of reliability analysis.
• Comparison of the three approaches reveals that:
1. Turkstra’s rule:
(+) Reliability can be assessed by any method for the time-
invariant analysis.
(+) Estimation of partial factors is straightforward.
(0) When applied strictly as proposed, failure probability may be
underestimated (error insignificant).
2. Ferry Borges-Castanheta models:
(+) The exact solution is found if time-variant loads are well
described by FBC models.
(-) Rackwitz-Fiessler algorithm may be unavailable in software.
(-) Estimation of partial factors may be complicated.
16
 Conclusions
3. Renewal processes:
(0) Estimation of partial factors is straightforward if a dominant
load case is identified.
(-) For dominant time-invariant variables, conservative results
are obtained.
(-) Upper bound on failure probability is not available in
software products.

• For common studies, Turkstra’s rule is recommended

(verification by FBC models).
• Renewal processes may be useful for non-stationary
conditions.

More details: Sýkora, M. - Holický, M. Comparison of load combination models for

probabilistic calibrations (to be published). In Proc. ICASP11, 1-4 August,
2011, ETH Zurich, Switzerland, 2011. 17
 Czech Technical University in Prague, Klokner Institute

Milan Holicky and Miroslav Sykora

miroslav.sykora@klok.cvut.cz
Competitive comparison of load combination models
Thank you for your attention.

References:
Ferry Borges & Castanheta (1971) ‘Structural Safety’, Course 101
(2nd ed.). Lisbon: Laboratorio National de Engenharia Civil.
Rackwitz (1998) ‘Computational Techniques in Stationary and Non-
Stationary Load Combination – A Review …’, J Structur Eng.
Sykora (2005) ‘Load Combination Model Based on Intermittent
Rectangular Wave Renewal Processes’, ICOSSAR 2005.
Turkstra (1970) ‘Theory of Structural Design Decisions’, University of
Waterloo, Ontario, Canada

1st International Symposium on Uncertainty Modelling in

Engineering, 2 - 3 May 2011, Prague, Czech Republic 18